Global Trends in Insurance Fraud : Next Generation Network Analytics

Transcription

1 SAS FINANCIAL CRIMES EXECUTIVE FORUM Toronto, 2018 Global Trends in Insurance Fraud : Next Generation Network Analytics David Hartley, Global Director SAS

2 Introduction Global Trends in Insurance Fraud

3 Global Insurance P&C Claims Fraud CAIF estimate $80bn losses annually. III estimates $32bn for P&C alone ABI estimates that undetected fraud cost insurers 3bn a year FFI 2014 survey - 19% said they knew a person who has deceived his/her insurance company. Svensk Försäkring estimate that 5-10% of claims include fraud and that 75% of people convicted for insurance fraud had previous criminal convictions and insurance fraud is more socially acceptable than other types of criminality DIA estimates that it costs the honest insurer holder up to DKK 500 per year (2013) ALFA estimate that fraud 15% of claims paid, or 4-8% of premiums collected equating to 2.5bn per annum GDV estimates that insurance fraud costs circa 4bn per annum VvV estimate that fraud has increased 25% in last 5 years adding 150 to a policy & that 10% of claims may be fraudulent ARIA estimates that 30% of compulsory motor, 20% of casco motor & 15% property claims are fraudulent Schweizerischer Versicherungsverband estimate that 10% of claims paid are fraudulent ICB estimate that insurance fraud inflates the cost of insurance by 15% IFBA estimates that insurance fraud costs more than $2 billion annually GIA of Singapore estimates are 20% of claims are fraudulent or overinflated at a cost of 140 M annually. Korean Financial Supervisory Service (FSS) estimates fraud losses $4.5 billion per year (13.5% in Life and 86.5% in Non-Life)

4 AVIVA Canada

5 Healthcare Fraud Global study says costs of fraud and error is higher than expected The research published in this report covers 14 different types of healthcare expenditure totalling $4.44 trillion (USD), in 33 organisations from 7 countries: UK, USA, France, Belgium, Netherlands, Australia & New Zealand Source: The Financial Cost of Healthcare Fraud 2015 University of Portsmouth Centre for Counter Fraud Studies & PKF Littlejohn LLP

6 FINDING THE FRAUD 100 MOTOR CLAIMS 100 Motor Claims

7 FINDING THE FRAUD THERE MAY BE 10 POTENTIAL FRAUDS CASES 100 Motor Claims 10 potential frauds cases

8 FINDING THE FRAUD WITH BUSINESS RULES THE INSURER MAY FIND 100 Motor Claims 10 potential frauds cases Business rules only

9 FINDING THE FRAUD LEAVING THE ANALYTICAL OPPORTUNITY 100 Motor Claims 10 potential frauds cases Business rules only

10 Amica Caixa esurance Chubb Grange HDI Qualitas Admiral Aksigorta Allianz Alm Brand Aviva Catalana Occ. Ethniki Generali ICB Ingosstrakh Lusitania Poste Vita SBM VIG Ydrogios ACC China Life Dongbu Hanwha Hyundai Nonghyup Kyobo IAG Ping An Zhongan Plus others under non-disclosure agreements Insurance Fraud Customers 90+

11 Large insurance company in Turkey Part owned by Ageas Aksigorta Turkey 66% increase in fraud detection since using the SAS solution Real time scores 8 seconds And making full use of social network analysis Proven & Captured Fraud Rates Based on # of claims 6.20% Proven & Captured Fraud Rates m 2.40% 3.00% 4.80%

12 CZ-Groep Netherlands Third largest healthcare insurer Moved from a pay and chase model to prepayment control Increased savings in first 12 months from 65m to 97m; +50% Now successfully rolled out across: Physiotherapy Mental health Pharmaceuticals Medical devises Next Stages Prevention is always better than cure. Through intelligent analysis, we can stay ahead of faulty statements and fraud. Marnix Suijkerbuijk Director of Health Care and Statement Service General practitioner Care Hospital Care Dental Full Story here

13 So How Can Analytics Help? Global Trends in Insurance Fraud

14 The Devil is in the Data

15 SAS Analytical Methodology - End to End Data Detection Investigation Reporting Structured & unstructured data sources Data cleansing Data integration Variable extraction & sentiment analysis with text mining Entity resolution Business rules Anomaly detection Advanced predictive models Watch lists Profiling Social network analysis and network-level analytics Automated alert generation Advanced ranking technology Custom alert queues Alert qualification and triage Powerful user interface with single and holistic views Dedicated dashboards, easy to use web-based interface Workflow analysis Full business intelligence reporting capability System and case management integration Dynamic data exploration Advanced query of integrated data Detection performance analysis New modus operandi discovery Discovery Accelerated design and constant improvement of the detection logic Alert suppression & routing rules Simulation and testing of new risk assessment methodologies

16 SAS Hybrid Scoring Approach for Fraud Levels Of Detection Event Machine Learning Entity Anomaly Detection Alert Generation Process Database Searches Network Automated Business Rules Text Mining

17

18 Fraud Detection The evolution of fraud detection solutions Initially fraud solutions were all based on simple binary rules Any item hitting on of the key risk factors would be sent for investigation and some of these would end up as fraud Rules Rule triggered? Investigation Fraud

19 Fraud Detection The evolution of fraud detection solutions Next advanced analytics was brought into the picture: Anomaly detection and text mining was used to discover new rules Predictive models were used to generate an overall scorecard using the historic outcomes Scorecard Score > threshold Investigation Fraud Use of analytical models to find new rules and produce weighted scorecard Fraud outcomes

20 Fraud Detection The evolution of fraud detection solutions Internal and external watchlists were then added based on both exact and fuzzy matching to further improve the scorecard Scorecard Score > threshold Investigation Fraud Fraud watchlist Use of analytical models to find new rules and produce weighted scorecard Fraud outcomes

21 Fraud Detection The evolution of fraud detection solutions For more complex cases some customers would create link diagrams. Sometimes this was done by hand on paper, other times tools like I2 were used to help to create these on a case by case basis Scorecard Score > threshold Investigation Complex case Link Analysis Fraud Fraud watchlist Use of analytical models to find new rules and produce weighted scorecard Fraud outcomes

22 Fraud Detection The evolution of fraud detection solutions In the final step, rather than creating the link charts manually for only high risk cases, new technology enabled the generation of these social networks automatically up front, enabling network level variables to be used in the score models and making them more accurate: Social Network Analysis Scorecard Score > threshold Investigation Link Analysis Fraud Fraud watchlist Use of analytical models to find new rules and produce weighted scorecard Fraud outcomes

23 Entities P&C Person - Party ID policyholder, claimant, supplier owner, insurance employee etc. Address - Insured address, risk address etc. Telephone numbers - Landline, Mobile (sequential) address IP address Suppliers garages/repair shop, medical, legal, hire car, tow truck etc. Insurance agent Bank account in/out

24 Entities Life Person - Party ID policyholder, beneficiary, insurance employee etc. Address - Insured address etc. Telephone numbers - Landline, Mobile (sequential) address IP address Suppliers medical service providers (disability, dread diseases) etc. Insurance agent Bank account in/out

25 Outpatient Rehabilitation Centers Podiatrists Air Transport Birthing Centers Entities ESRD Treatment Facilities Health - Providers Imaging Centers Occupational Therapist Trauma Centers Ambulatory Surgical Centers Doctors Medical Equipment Suppliers Substance Abuse Facilities Dentists Hospitals Retail Health Outlet Mobile Health Urgent Care Facilities Pharmacies Chiropractors Dialysis Centers Land Transport Optometrists Infusion Centers Hospice Centers Psychiatric Facilities Home Health Care Laboratories Substance Abuse Facilities Therapeutic Massage Therapists Physical Therapists Personal Care Assistants Inpatient Rehabilitation Centers Social Worker Rural Health Intermediate Pathology

26 Social Networks Visualising Unexplained Relationships The high level process for producing social networks is as follows: Documents Elements Compounds Entities Networks Documents: Often best thought of as what you might find on a piece of paper these contain all of the data relate to a business understood concept. E.g. Insurance Claim Insurance policy application Elements: fields within the incoming data that partially or completely identify an entity. E.g. Forename Surname Date of Birth Social Security # Telephone number Compounds: An understanding of the real world is used to combine elements into combinations that represent real-world items within the data. E.g: Fore+Sur+DoB Fore+Sur+Hse#+Street+Zip Social Security # Entities: The realworld items that we are trying to model within our fraud solution. E.g. Individual Address Landline Bank Account Address Networks: Groups of strongly connected documents and entities IP Address

27

28

29 Agent Gaming Social Network Analysis Example: 5 written off policies connected to the same phone number

30 Internal Fraud Network Analysis

31 The Challenge

32 The Challenge How to identify a potential organised crime ring in less than 10 minutes

33 sas.com